All showcase images have WF embedded.<h2 id="trained-on-this-type-of-prompt">Trained on this type of prompt</h2>Headscissors: a (digital camera snapshot / phone snapshot / video frame grab...) of a woman doing a headscissors on a man...Reverse Headscissors: a (digital camera snapshot / phone snapshot / video frame grab...) of a woman doing a reverse headscissors on a man...<h2 id="recommended-settings">Recommended settings</h2><ul><li>Sampler: euler_ancestral</li><li>Scheduler: Beta57</li><li>Steps: 8-12</li><li>LoRA strength 1.0 (feel free to experiment)</li><li>Combine with <code>Krea2_TextFusion_Refusal_Reduction</code> at 1.0 strength for better prompt adherence</li><li>For NSFW elements, I also combine it with <code>KNP_000003000</code> at 0.6 strenght</li></ul>Note that these settings are experimental and based on my tests. There might be better combinations and LoRA strengths.<h2 id="known-limitations-(honest-list)">Known limitations (honest list)</h2><ul><li>You may need to adapt your prompt to clearly distinguish between reverse headscissors and regular headscissors. The model can struggle to tell the difference between reverse and classic headscissors. It’s recommended to describe the overall position in natural language to avoid the “casino effect,” as the LoRA was trained on various headscissors poses</li><li>Sometimes feet or hands can get messed up. You can try another seed, sampler, or resolution</li><li>The LoRA doesn’t have a great understanding of male anatomy. If a man is naked, try adding an NSFW LoRA or a penis helper to achieve better realism</li></ul>

Files with SHA256: f5483054935751e2417d4864eb10d450b8d1b207accb23fe6a63baffa160bb1d